[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-117821-en":3,"doc-seo-117821-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},117821,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Modular Lifelong Machine Learning - Doctor of Philosophy Thesis","Deep learning accelerates progress across domains such as computer vision and natural language processing, but training large networks is costly and becomes even more demanding when solving additional tasks. Lifelong machine learning addresses this by learning a stream of problems as they arrive and transferring previously acquired knowledge, while still maintaining strong performance across all encountered tasks to avoid catastrophic forgetting. This thesis proposes modular approaches that improve transfer properties and scalability, introducing HOUDINI, PICLE, and a multi-fidelity black-box optimization method to support longer sequences.","This thesis has been submitted in fulfilment of the requirements for a postgraduate degree (e. g. PhD, MPhil, DClinPsychol) at the University of Edinburgh. Please note the following terms and conditions of use:  \n• This work is protected by copyright and other intellectual property rights, which are retained by the thesis author, unless otherwise stated.  \n• A copy can be downloaded for personal non-commercial research or study, without prior permission or charge.  \n• This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the author.  \n• The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the author.  \n• When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given.  \nModular Lifelong Machine Learning  \nLazar Ignatov Valkov  \nU  \nR  \nG  \nH  \nO  \nF  \nE  \nD  \nDoctor of Philosophy  \nInstitute for Adaptive and Neural Computation School of Informatics  \nUniversity of Edinburgh  \nAbstract  \nDeep learning has drastically improved the state-of-the-art in many important fields, including computer vision and natural language processing (LeCun et al., 2015) . However, it is expensive to train a deep neural network on a machine learning problem. The overall training cost further increases when one wants to solve additional problems. Lifelong machine learning (LML) develops algorithms that aim to efficiently learn to solve a sequence of problems, which become available one at a time. New problems are solved with less resources by transferring previously learned knowledge. At the same time, an LML algorithm needs to retain good performance on all encountered problems, thus avoiding catastrophic forgetting. Current approaches do not possess all the desired properties of an LML algorithm. First, they primarily focus on preventing catastrophic forgetting (D´ıaz-Rodr´ıguez et al., 2018; Delange et al., 2021) . As a result, they neglect some knowledge transfer properties. Furthermore, they assume that all problems in a sequence share the same input space. Finally, scaling these methods toa large sequence of problems remains a challenge.  \nModular approaches to deep learning decompose a deep neural network into subnetworks, referred to as modules. Each module can then be trained to perform anatomic transformation, specialised in processing a distinct subset of inputs. This modular approach to storing knowledge makes it easy to only reuse the subset of modules which are useful for the task at hand.  \nThis thesis introduces a line of research which demonstrates the merits of a modular approach to lifelong machine learning, and its ability to address the aforementioned shortcomings of other methods. Compared to previous work, we show that a modular approach can be used to achieve more LML properties than previously demonstrated. Furthermore, we develop tools which allow modular LML algorithms to scale in order to retain said properties on longer sequences of problems.  \nFirst, we introduce HOUDINI, a neurosymbolic framework for modular LML. HOUDINI represents modular deep neural networks as functional programs and accumulatesa library of pre-trained modules over a sequence of problems. Given a new problem, we use program synthesis to select a suitable neural architecture, as well as a highperforming combination of pre-trained and new modules. We show that our approach has most of the properties desired from an LML algorithm. Notably, it can perform forward transfer, avoid negative transfer and prevent catastrophic forgetting, even across problems with disparate input domains and problems which require different neural architectures.  \nSecond, we produce a modular LML algorithm which retains the properties of HOUDINI but can also scale to longer sequences of problems. To this end, we fix the choice of a neural architecture and introduce a probabili","cbCaiqRsCLkl2ptK","https://ap.wps.com/l/cbCaiqRsCLkl2ptK","pdf",3395476,1,208,"English","en",105,"# Abstract\n# Lay Summary\n## Modular approach to lifelong learning\n## HOUDINI framework\n## PICLE for longer sequences\n## Black-box optimization and multi-fidelity HPO\n# Overall contributions","[{\"question\":\"What problem does lifelong machine learning address in this thesis?\",\"answer\":\"It addresses efficient learning over a sequence of problems that arrive one at a time, using knowledge transfer to reduce resources while preventing catastrophic forgetting.\"},{\"question\":\"What is HOUDINI and how does it support modular lifelong learning?\",\"answer\":\"HOUDINI represents modular deep networks as functional programs and maintains a library of pre-trained modules. For a new problem, program synthesis selects an architecture and combines pre-trained and newly trained modules.\"},{\"question\":\"How does PICLE help modular lifelong machine learning scale to longer sequences?\",\"answer\":\"PICLE fixes the neural architecture choice and introduces a probabilistic search framework to find strong module combinations efficiently, supported by probabilistic models over modules.\"}]","Modular Lifelong Machine Learning - Doctor of Philosophy Thesis | PDF",1785679790,524,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"modular-lifelong-machine-learning-doctor-of-philosophy-thesis","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/modular-lifelong-machine-learning-doctor-of-philosophy-thesis/117821/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-05","2026-08-02",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does lifelong machine learning address in this thesis?","Question",{"text":76,"@type":77},"It addresses efficient learning over a sequence of problems that arrive one at a time, using knowledge transfer to reduce resources while preventing catastrophic forgetting.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What is HOUDINI and how does it support modular lifelong learning?",{"text":81,"@type":77},"HOUDINI represents modular deep networks as functional programs and maintains a library of pre-trained modules. For a new problem, program synthesis selects an architecture and combines pre-trained and newly trained modules.",{"name":83,"@type":74,"acceptedAnswer":84},"How does PICLE help modular lifelong machine learning scale to longer sequences?",{"text":85,"@type":77},"PICLE fixes the neural architecture choice and introduces a probabilistic search framework to find strong module combinations efficiently, supported by probabilistic models over modules.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,121,124,129,132,136],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":46,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":133,"doc_module":4,"doc_module_name":46,"category_name":134,"show_sort_weight":133,"slug":135},10,"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":46,"category_name":138,"show_sort_weight":107,"slug":139},19,"General","general"]